From Anadoris. This guide comes from the team behind Digital & Technology: websites, ERP, CRM and integrations, built by people who run businesses. About the group.
Separate three different questions
Freshness asks how recently the relevant data was updated. Coverage asks whether the expected feeds, periods or records arrived. Correctness asks whether those records produce the expected meaning and values. Google’s SRE guidance treats these as distinct concerns for data-processing pipelines.
For a practical business example, imagine an order feed arrived this morning but the returns feed did not. The dashboard may be fresh in one sense while overstating the information available for a margin decision. A single green timestamp should not hide that distinction. List the feeds required for each important business calculation.
Define usable data for each decision
Write a short acceptance rule for the decision, not merely the integration. A stock report might require the latest accepted warehouse movement data and a reconciliation of important locations. A contribution view might require costs, fees and returns for the relevant period. Document which inputs are estimates and how the screen should identify them.
Decide how the system behaves when an input fails. Options include showing the last validated value with its date, marking the calculation incomplete or blocking the affected automated action. Choose deliberately. Quietly substituting a missing value with zero can make a report look more certain than the evidence allows.
Build an exception view people can act on
Show the affected feed, last successful validation, missing scope, business impact and named owner. Keep technical details accessible without making every user interpret raw logs. A finance colleague should be able to see which decision needs caution even if the underlying failure is an API issue.
Start with one critical calculation and trace it backwards to its inputs. Test the display by deliberately withholding a non-production sample feed and checking that the result is marked appropriately. This is a proposed control pattern, not a claim that every Anadoris system already implements it. Expand only after the first workflow makes failures visible and recovery clear.
Sources
- {'title': 'Google SRE: Data Processing Pipelines', 'url': 'https://sre.google/workbook/data-processing/'}
